Key Takeaways
1. Nvidia announced native support for CUDA on RISC-V, highlighting its potential as a competitor to x86 and ARM architectures.
2. The announcement was made by Nvidia’s VP, Frans Sijstermans, at a RISC-V summit in China, showcasing how RISC-V can manage CUDA drivers.
3. The integration of CUDA with RISC-V may help Nvidia expand its ecosystem in China, where RISC-V is growing rapidly.
4. AMD is developing its own alternative to CUDA, called ROCm, which has also added support for RISC-V, indicating competition in the market.
5. The adoption of ROCm is slow, suggesting it may take time before there is significant competition against Nvidia’s CUDA in computing software.
Back in 2021, researchers focusing on RISC-V showed that Nvidia’s CUDA code could operate on non-proprietary hardware like a Vortex GPGPU that is based on RISC-V, utilizing an OpenCL translator. While this might not be the most effective method to implement CUDA support for RISC-V, it clearly indicated a rising interest in RISC-V as a serious competitor to x86 and ARM processing architectures. Nvidia is now formally recognizing the potential of RISC-V in the computing field by announcing the native support of CUDA for RISC-V.
Significant Announcement
The announcement was made by Frans Sijstermans, Nvidia’s Vice President of Hardware Engineering, during a RISC-V summit held in China. Sijstermans is also a director on the RISC-V board. A diagram showcased during the event illustrated how the RISC-V processor can manage CUDA drivers at the operating system level, while Nvidia’s GPUs run the CUDA kernels. There is a Data Processing Unit (DPU) mentioned, presumably from Nvidia as well, which indicates that the diagram depicts a computing system for high-performance computing (HPC) and data centers.
Industry Reactions
According to Tom’s Hardware, this step to provide CUDA support for open-source architectures like RISC-V might allow Team Green to broaden its ecosystem in China, where RISC-V is experiencing rapid growth, despite the limitations on marketing AI accelerators like the GB200 and GB300 in that area.
In contrast, AMD is pushing its own alternative to Nvidia’s CUDA through ROCm, which is now in its seventh version and has already added support for the RISC-V architecture. Numerous companies have voiced their opposition to the dominance of CUDA, and Team Red is making efforts to challenge the monopoly. However, the adoption of ROCm remains gradual, and it may take a few more years before we witness true competition in the realm of computing software.
Source:
Link
Leave a Reply